idnits 2.17.1 draft-ietf-nvo3-mcast-framework-01.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- == There are 1 instance of lines with multicast IPv4 addresses in the document. If these are generic example addresses, they should be changed to use the 233.252.0.x range defined in RFC 5771 Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (November 9, 2015) is 3090 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- == Missing Reference: 'VXLAN' is mentioned on line 164, but not defined == Missing Reference: 'FW' is mentioned on line 181, but not defined == Unused Reference: 'RFC7348' is defined on line 514, but no explicit reference was found in the text Summary: 0 errors (**), 0 flaws (~~), 5 warnings (==), 1 comment (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 1 NVO3 working group A. Ghanwani 2 Internet Draft Dell 3 Intended status: Informational L. Dunbar 4 Expires: May 9, 2016 M. McBride 5 Huawei 6 V. Bannai 7 Google 8 R. Krishnan 9 Dell 11 November 9, 2015 13 A Framework for Multicast in NVO3 14 draft-ietf-nvo3-mcast-framework-01 16 Status of this Memo 18 This Internet-Draft is submitted in full conformance with the 19 provisions of BCP 78 and BCP 79. 21 This Internet-Draft is submitted in full conformance with the 22 provisions of BCP 78 and BCP 79. This document may not be modified, 23 and derivative works of it may not be created, except to publish it 24 as an RFC and to translate it into languages other than English. 26 Internet-Drafts are working documents of the Internet Engineering 27 Task Force (IETF), its areas, and its working groups. Note that 28 other groups may also distribute working documents as Internet- 29 Drafts. 31 Internet-Drafts are draft documents valid for a maximum of six 32 months and may be updated, replaced, or obsoleted by other documents 33 at any time. It is inappropriate to use Internet-Drafts as 34 reference material or to cite them other than as "work in progress." 36 The list of current Internet-Drafts can be accessed at 37 http://www.ietf.org/ietf/1id-abstracts.txt 39 The list of Internet-Draft Shadow Directories can be accessed at 40 http://www.ietf.org/shadow.html 42 This Internet-Draft will expire on May 9, 2016. 44 Copyright Notice 46 Copyright (c) 2015 IETF Trust and the persons identified as the 47 document authors. All rights reserved. 49 This document is subject to BCP 78 and the IETF Trust's Legal 50 Provisions Relating to IETF Documents 51 (http://trustee.ietf.org/license-info) in effect on the date of 52 publication of this document. Please review these documents 53 carefully, as they describe your rights and restrictions with 54 respect to this document. Code Components extracted from this 55 document must include Simplified BSD License text as described in 56 Section 4.e of the Trust Legal Provisions and are provided without 57 warranty as described in the Simplified BSD License. 59 Abstract 61 This document discusses a framework of supporting multicast traffic 62 in a network that uses Network Virtualization Overlays over Layer 3 63 (NVO3). Both infrastructure multicast and application-specific 64 multicast are discussed. It describes the various mechanisms that 65 can be used for delivering such traffic as well as the data plane 66 and control plane considerations for each of the mechanisms. 68 Table of Contents 70 1. Introduction...................................................3 71 2. Acronyms.......................................................4 72 3. Multicast mechanisms in networks that use NVO3.................4 73 3.1. No multicast support......................................5 74 3.2. Replication at the source NVE.............................6 75 3.3. Replication at a multicast service node...................8 76 3.4. IP multicast in the underlay..............................9 77 3.5. Other schemes............................................10 78 4. Simultaneous use of more than one mechanism...................10 79 5. Other issues..................................................11 80 5.1. Multicast-agnostic NVEs..................................11 81 5.2. Multicast membership management for DC with VMs..........12 82 6. Summary.......................................................12 83 7. Security Considerations.......................................12 84 8. IANA Considerations...........................................12 85 9. References....................................................12 86 9.1. Normative References.....................................12 87 9.2. Informative References...................................13 89 10. Acknowledgments..............................................14 91 1. Introduction 93 Network virtualization using Overlays over Layer 3 (NVO3) is a 94 technology that is used to address issues that arise in building 95 large, multitenant data centers that make extensive use of server 96 virtualization [RFC7364]. 98 This document provides a framework for supporting multicast traffic, 99 in a network that uses Network Virtualization using Overlays over 100 Layer 3 (NVO3). Both infrastructure multicast (ARP/ND, DHCP, mDNS, 101 etc.) and application-specific multicast are considered. It 102 describes the various mechanisms and considerations that can be used 103 for delivering such traffic in networks that use NVO3. 105 The reader is assumed to be familiar with the terminology as defined 106 in the NVO3 Framework document [RFC7365] and NVO3 Architecture 107 document [NVO3-ARCH]. 109 1.1. Infrastructure multicast 111 Infrastructure multicast includes protocols such as ARP/ND, DHCP, 112 and mDNS. It is possible to provide solutions for these that do not 113 involve multicast in the underlay network. In the case of ARP/ND, 114 an NVA can be used for distributing the mappings of IP address to 115 MAC address to all NVEs, and the NVEs can respond to ARP messages 116 from the TSs that are attached to it in a way that is similar to 117 proxy-ARP. In the case of DHCP, the NVE can be configured to 118 forward these messages using a helper function. 120 Of course it is possible to support all of these infrastructure 121 multicast protocols natively if the underlay provides multicast 122 transport. However, even in the presence of multicast transport, it 123 may be beneficial to use the optimizations mentioned above to reduce 124 the amount of such traffic in the network. 126 1.2. Application-specific multicast 128 Application-specific multicast traffic, which may be either Source- 129 Specific Multicast (SSM) or Any-Source Multicast (ASM)[RFC 3569], 130 has the following characteristics: 132 1. Receiver hosts are expected to subscribe to multicast content 133 using protocols such as IGMP [RFC3376] (IPv4) or MLD (IPv6). 135 Multicast sources and listeners participant in these protocols 136 using addresses that are in the Tenant System address domain. 138 2. The list of multicast listeners for each multicast group is not 139 known in advance. Therefore, it may not be possible for an NVA 140 to get the list of participants for each multicast group ahead 141 of time. 143 2. Acronyms 145 ASM: Any-Source Multicast 147 LISP: Locator/ID Separation Protocol 149 NVA: Network Virtualization Authority 151 NVE: Network Virtualization Edge 153 NVGRE: Network Virtualization using GRE 155 SSM: Source-Specific Multicast 157 STT: Stateless Tunnel Transport 159 VXLAN: Virtual eXtensible LAN 161 3. Multicast mechanisms in networks that use NVO3 163 In NVO3 environments, traffic between NVEs is transported using an 164 encapsulation such as VXLAN [VXLAN], NVGRE [NVGRE], STT [STT], etc. 166 Besides the need to support the Address Resolution Protocol (ARP) 167 and Neighbor Discovery (ND), there are several applications that 168 require the support of multicast and/or broadcast in data centers 169 [DC-MC]. With NVO3, there are many possible ways that multicast may 170 be handled in such networks. We discuss some of the attributes of 171 the following four methods: 173 1. No multicast support. 175 2. Replication at the source NVE. 177 3. Replication at a multicast service node. 179 4. IP multicast in the underlay. 181 These mechanisms are briefly mentioned in the NVO3 Framework [FW] 182 and NVO3 architecture [NVO3-ARCH] document. This document attempts 183 to provide more details about the basic mechanisms underlying each 184 of these mechanisms and discusses the issues and tradeoffs of each. 186 We note that other methods are also possible, such as [EDGE-REP], 187 but we focus on the above four because they are the most common. 189 3.1. No multicast support 191 In this scenario, there is no support whatsoever for multicast 192 traffic when using the overlay. This method can only work if the 193 following conditions are met: 195 1. All of the application traffic in the network is unicast 196 traffic and the only multicast/broadcast traffic is from ARP/ND 197 protocols. 199 2. A network virtualization authority (NVA) is used by the NVEs to 200 determine the mapping of a given Tenant System's MAC/IP address 201 to its NVE. In other words, there is no data plane learning. 202 Address resolution requests via ARP/ND that are issued by the 203 Tenant Systems must be resolved by the NVE that they are 204 attached to. 206 With this approach, it is not possible to support application- 207 specific multicast. However, certain multicast/broadcast 208 applications such as DHCP can be supported by use of a helper 209 function in the NVE. 211 The main drawback of this approach, even for unicast traffic, is 212 that it is not possible to initiate communication with a Tenant 213 System for which a mapping to an NVE does not already exist with the 214 NVA. This is a problem in the case where the NVE is implemented in 215 a physical switch and the Tenant System is a physical end station 216 that has not registered with the NVA. 218 3.2. Replication at the source NVE 220 With this method, the overlay attempts to provide a multicast 221 service without requiring any specific support from the underlay, 222 other than that of a unicast service. A multicast or broadcast 223 transmission is achieved by replicating the packet at the source 224 NVE, and making copies, one for each destination NVE that the 225 multicast packet must be sent to. 227 For this mechanism to work, the source NVE must know, a priori, the 228 IP addresses of all destination NVEs that need to receive the 229 packet. For the purpose of ARP/ND, this would involve knowing the 230 IP addresses of all the NVEs that have Tenant Systems in the virtual 231 network instance (VNI) of the Tenant System that generated the 232 request. For the support of application-specific multicast traffic, 233 a method similar to that of receiver-sites registration for a 234 particular multicast group described in [LISP-Signal-Free] can be 235 used. The registrations from different receiver-sites can be merged 236 at the NVA, which can construct a multicast replication-list 237 inclusive of all NVEs to which receivers for a particular multicast 238 group are attached. The replication-list for each specific multicast 239 group is maintained either by the NVA. 241 The receiver-sites registration is achieved by egress NVEs 242 performing the IGMP/MLD snooping to maintain state for which 243 attached Tenant Systems have subscribed to a given IP multicast 244 group. When the members of a multicast group are outside the NVO3 245 domain, it is necessary for NVO3 gateways to keep track of the 246 remote members of each multicast group. The NVEs then communicate 247 these mappings to the NVA. Even if the membership is not 248 communicated to the NVA, if it is necessary to prevent hosts 249 attached to an NVE that have not subscribed to a multicast group 250 from receiving the multicast traffic, the NVE needs to maintain the 251 multicast group membership. 253 In the absence of IGMP/MLD snooping, the traffic would be delivered 254 to all hosts that are part of the VNI. 256 This method requires multiple copies of the same packet to all NVEs 257 that participate in the VN. If, for example, a tenant subnet is 258 spread across 50 NVEs, the packet would have to be replicated 50 259 times at the source NVE. This also creates an issue with the 260 forwarding performance of the NVE. 262 Note that this method is similar to what was used in VPLS [VPLS] 263 prior to support of MPLS multicast [MPLS-MC]. While there are some 264 similarities between MPLS VPN and the NVO3 overlay, there are some 265 key differences: 267 - The CE-to-PE attachment in VPNs is somewhat static, whereas in a 268 DC that allows VMs to migrate anywhere, the TS attachment to NVE 269 is much more dynamic. 271 - The number of PEs to which a single VPN customer is attached in 272 an MPLS VPN environment is normally far less than the number of 273 NVEs to which a VNI's VMs are attached in a DC. 275 When a VPN customer has multiple multicast groups, [RFC6513] 276 "Multicast VPN" combines all those multicast groups within each 277 VPN client to one single multicast group in the MPLS (or VPN) 278 core. The result is that messages from any of the multicast 279 groups belonging to one VPN customer will reach all the PE nodes 280 of the client. In other words, any messages belonging to any 281 multicast groups under customer X will reach all PEs of the 282 customer X. When the customer X is attached to only a handful of 283 PEs, the use of this approach does not result in excessive wastage 284 of bandwidth in the provider's network. 286 In a DC environment, a typical server/hypervisor based virtual 287 switch may only support 10's VMs (as of this writing). A subnet 288 with N VMs may be, in the worst case, spread across N vSwitches. 289 Using "MPLS VPN multicast" approach in such a scenario would 290 require the creation of a Multicast group in the core for this VNI 291 to reach all N NVEs. If only small percentage of this client's VMs 292 participate in application specific multicast, a great number of 293 NVEs will receive multicast traffic that is not forwarded to any 294 of their attached VMs, resulting in considerable wastage of 295 bandwidth. 297 Therefore, the Multicast VPN solution may not scale in DC 298 environment with dynamic attachment of Virtual Networks to NVEs and 299 greater number of NVEs for each virtual network. 301 3.3. Replication at a multicast service node 303 With this method, all multicast packets would be sent using a 304 unicast tunnel encapsulation to a multicast service node (MSN). The 305 MSN, in turn, would create multiple copies of the packet and would 306 deliver a copy, using a unicast tunnel encapsulation, to each of the 307 NVEs that are part of the multicast group for which the packet is 308 intended. 310 This mechanism is similar to that used by the ATM Forum's LAN 311 Emulation [LANE] specification [LANE]. 313 The following are the possible ways for the MSN to get the 314 membership information for each multicast group: 316 - The MSN can obtain this information by snooping the IGMP/MLD 317 messages from the Tenant Systems and/or sending query messages to 318 the Tenant Systems. In order for MSN to snoop the IGMP/MLD 319 messages between TSs and their corresponding routers, the NVEs 320 that TSs are attached have to encapsulate a special outer header, 321 e.g. outer destination being the multicast server node. See 322 Section 3.3.2 for detail. 324 - The MSN can obtain the membership information from the NVEs that 325 snoop the IGMP/MLD messages. This can be done by having the MSN 326 communicate with the NVEs, or by having the NVA obtain the 327 information from the NVEs, and in turn have MSN communicate with 328 the NVA. 330 Unlike the method described in Section 3.2, there is no performance 331 impact at the ingress NVE, nor are there any issues with multiple 332 copies of the same packet from the source NVE to the multicast 333 service node. However there remain issues with multiple copies of 334 the same packet on links that are common to the paths from the MSN 335 to each of the egress NVEs. Additional issues that are introduced 336 with this method include the availability of the MSN, methods to 337 scale the services offered by the MSN, and the sub-optimality of the 338 delivery paths. 340 Finally, the IP address of the source NVE must be preserved in 341 packet copies created at the multicast service node if data plane 342 learning is in use. This could create problems if IP source address 343 reverse path forwarding (RPF) checks are in use. 345 3.4. IP multicast in the underlay 347 In this method, the underlay supports IP multicast and the ingress 348 NVE encapsulates the packet with the appropriate IP multicast 349 address in the tunnel encapsulation header for delivery to the 350 desired set of NVEs. The protocol in the underlay could be any 351 variant of Protocol Independent Multicast (PIM), or protocol 352 dependent multicast, such as [ISIS-Multicast]. 354 If an NVE connects to its attached TSs via Layer 2 network, there 355 are multiple ways for NVEs to support the application specific 356 multicast: 358 - The NVE only supports the basic IGMP/MLD snooping function, let 359 the TSs routers handling the application specific multicast. This 360 scheme doesn't utilize the underlay IP multicast protocols. 361 - 362 - The NVE can act as a pseudo multicast router for the directly 363 attached VMs and support proper mapping of IGMP/MLD's messages to 364 the messages needed by the underlay IP multicast protocols. 366 With this method, there are none of the issues with the methods 367 described in Sections 3.2. 369 With PIM Sparse Mode (PIM-SM), the number of flows required would be 370 (n*g), where n is the number of source NVEs that source packets for 371 the group, and g is the number of groups. Bidirectional PIM (BIDIR- 372 PIM) would offer better scalability with the number of flows 373 required being g. 375 In the absence of any additional mechanism, e.g. using an NVA for 376 address resolution, for optimal delivery, there would have to be a 377 separate group for each tenant, plus a separate group for each 378 multicast address (used for multicast applications) within a tenant. 380 Additional considerations are that only the lower 23 bits of the IP 381 address (regardless of whether IPv4 or IPv6 is in use) are mapped to 382 the outer MAC address, and if there is equipment that prunes 383 multicasts at Layer 2, there will be some aliasing. Finally, a 384 mechanism to efficiently provision such addresses for each group 385 would be required. 387 There are additional optimizations which are possible, but they come 388 with their own restrictions. For example, a set of tenants may be 389 restricted to some subset of NVEs and they could all share the same 390 outer IP multicast group address. This however introduces a problem 391 of sub-optimal delivery (even if a particular tenant within the 392 group of tenants doesn't have a presence on one of the NVEs which 393 another one does, the former's multicast packets would still be 394 delivered to that NVE). It also introduces an additional network 395 management burden to optimize which tenants should be part of the 396 same tenant group (based on the NVEs they share), which somewhat 397 dilutes the value proposition of NVO3 which is to completely 398 decouple the overlay and physical network design allowing complete 399 freedom of placement of VMs anywhere within the data center. 401 Multicast schemes such as BIER (Bit Index Explicit Replication) may 402 be able to provide optimizations by allowing the underlay network to 403 provide optimum multicast delivery without requiring routers in the 404 core of the network to main per-multicast group state. 406 3.5. Other schemes 408 There are still other mechanisms that may be used that attempt to 409 combine some of the advantages of the above methods by offering 410 multiple replication points, each with a limited degree of 411 replication [EDGE-REP]. Such schemes offer a trade-off between the 412 amount of replication at an intermediate node (router) versus 413 performing all of the replication at the source NVE or all of the 414 replication at a multicast service node. 416 4. Simultaneous use of more than one mechanism 418 While the mechanisms discussed in the previous section have been 419 discussed individually, it is possible for implementations to rely 420 on more than one of these. For example, the method of Section 3.1 421 could be used for minimizing ARP/ND, while at the same time, 422 multicast applications may be supported by one, or a combination of, 423 the other methods. For small multicast groups, the methods of 424 source NVE replication or the use of a multicast service node may be 425 attractive, while for larger multicast groups, the use of multicast 426 in the underlay may be preferable. 428 5. Other issues 430 5.1. Multicast-agnostic NVEs 432 Some hypervisor-based NVEs do not process or recognize IGMP/MLD 433 frames; i.e. those NVEs simply encapsulate the IGMP/MLD messages in 434 the same way as they do for regular data frames. 436 By default, TSs router periodically sends IGMP/MLD query messages to 437 all the hosts in the subnet to trigger the hosts that are interested 438 in the multicast stream to send back IGMP/MLD reports. In order for 439 MSN get the updated multicast group information, the MSN can also 440 send the IGMP/MLD query message comprising a client specific 441 multicast address, encapsulated in an overlay header to all the NVEs 442 to which the TSs in the VN are attached. 444 However, MSN may not always be aware of the client specific 445 multicast addresses. Then MSN has to snoop the IGMP/MLD messages 446 between TSs and their corresponding routers to maintain the 447 multicast membership. In order for MSN to snoop the IGMP/MLD 448 messages between TSs and their router, NVA needs to configure the 449 NVE to send copies of the IGMP/MLD messages to the MSN in addition 450 to the default behavior of sending them to the TSs' routers; e.g. 451 the NVA has to inform the NVEs to encapsulate data frames with DA 452 being 224.0.0.2 (destination address of IGMP report) to TSs' router 453 and MSN. 455 This process is similar to "Source Replication" described in Section 456 3.2, except the NVEs only replicate the message to TS's router and 457 MSN. 459 5.2. Multicast membership management for DC with VMs 461 For data centers with virtualized servers, VMs can be added, deleted 462 or moved very easily. When VMs are added, deleted or moved, the NVEs 463 to which the VMs are attached are changed. 465 When a VM is deleted from an NVE or a new VM is added to an NVE, the 466 VM management system should notify the MSN to send the IGMP/MLD 467 query messages to the relevant NVEs, so that the multicast 468 membership can be updated promptly. Otherwise, if there are changes 469 of VMs attachment to NVEs, then for the duration of the configured 470 default time interval that the TSs routers use for IGMP/MLD queries, 471 multicast data may not reach the VM(s) that moved. 473 6. Summary 475 This document has identified various mechanisms for supporting 476 application specific multicast in networks that use NVO3. It 477 highlights the basics of each mechanism and some of the issues with 478 them. As solutions are developed, the protocols would need to 479 consider the use of these mechanisms and co-existence may be a 480 consideration. It also highlights some of the requirements for 481 supporting multicast applications in an NVO3 network. 483 7. Security Considerations 485 This draft does not introduce any new security considerations beyond 486 what may be present in proposed solutions 488 8. IANA Considerations 490 This document requires no IANA actions. RFC Editor: Please remove 491 this section before publication. 493 9. References 495 9.1. Normative References 497 [RFC7365] Lasserre, M. et al., "Framework for data center (DC) 498 network virtualization", October 2014. 500 [RFC7364] Narten, T. et al., "Problem statement: Overlays for 501 network virtualization", October 2014. 503 [NVO3-ARCH] Narten, T. et al.," An Architecture for Overlay Networks 504 (NVO3)", work in progress, February 2014. 506 [RFC3376] B. Cain, et al, "Internet Group Management Protocol, 507 Version 3", October 2002. 509 [RFC6513] Rosen, E. et al., "Multicast in MPLS/BGP IP VPNs", 510 February 2012. 512 9.2. Informative References 514 [RFC7348] Mahalingam, M. et al., " Virtual eXtensible Local Area 515 Network (VXLAN): A Framework for Overlaying Virtualized 516 Layer 2 Networks over Layer 3 Networks", August 2014. 518 [NVGRE] Sridharan, M. et al., "NVGRE: Network virtualization using 519 Generic Routing Encapsulation", work in progress. 521 [STT] Davie, B. and Gross J., "A stateless transport tunneling 522 protocol for network virtualization," work in progress. 524 [DC-MC] McBride M., and Lui, H., "Multicast in the data center 525 overview," work in progress. 527 [ISIS-Multicast] 529 L. Yong, et al, "ISIS Protocol Extension For Building 530 Distribution Trees", work in progress. Oct 2013. 532 [VPLS] Lasserre, M., and Kompella, V. (Eds), "Virtual Private LAN 533 Service (VPLS) using Label Distribution Protocol (LDP) 534 signaling," RFC 4762, January 2007. 536 [MPLS-MC] Aggarwal, R. et al., "Multicast in VPLS," work in 537 progress. 539 [LANE] "LAN emulation over ATM," The ATM Forum, af-lane-0021.000, 540 January 1995. 542 [EDGE-REP] 543 Marques P. et al., "Edge multicast replication for BGP IP 544 VPNs," work in progress, June 2012. 546 [RFC 3569] 548 S. Bhattacharyya, Ed., "An Overview of Source-Specific 549 Multicast (SSM)", July 2003. 551 [LISP-Signal-Free] 553 V. Moreno & D. Farinacci, "Signal-Free LISP Multicast", 554 work in progress. Dec 2014. 556 10. Acknowledgments 558 Thanks are due to Dino Farinacci and Erik Nordmark for their 559 comments and suggestions. 561 This document was prepared using 2-Word-v2.0.template.dot. 563 Authors' Addresses 565 Anoop Ghanwani 566 Dell 567 Email: anoop@alumni.duke.edu 569 Linda Dunbar 570 Huawei Technologies 571 5340 Legacy Drive, Suite 1750 572 Plano, TX 75024, USA 573 Phone: (469) 277 5840 574 Email: ldunbar@huawei.com 576 Mike McBride 577 Huawei Technologies 578 mmcbride7@gmail.com 580 Vinay Bannai 581 Google 582 Email: vbannai@gmail.com 584 Ram Krishnan 585 Dell 586 Email: Ramki_Krishnan@dell.com